fuse hosts list

List all registered hosts, or select an active one

List

fuse hosts
fuse hosts list

Aliases on the parent: host-list. list alias: ls. Flags: --limit (page size, default 50, max 200), --cursor (resume from a previous page) and --all-pages (fetch every host). Both forms are equivalent, fuse hosts with no subcommand runs list.

$ fuse hosts
╭───┬─────────────┬─────────┬────────┬───────┬──────────────┬────────────┬──────┬──────────────────────────┬───────────╮
│   │ ID          │ REGION  │ STATE  │ CPUS  │ RAM MB       │ GPUS       │ VMS  │ FIRECRACKER              │ LAST SEEN │
├───┼─────────────┼─────────┼────────┼───────┼──────────────┼────────────┼──────┼──────────────────────────┼───────────┤
│ * │ prod-east-1 │ us-east │ active │ 4/16  │ 8192/65536   │ -          │ 1/20 │ v1.17.0                  │ 3m        │
│   │ prod-west-1 │ us-west │ active │ 0/16  │ 0/65536      │ -          │ 0/20 │ v1.16.0 (pinned v1.17.0) │ 1h        │
│   │ gpu-1       │ us-east │ active │ 16/64 │ 65536/262144 │ 2/4 (a100) │ 2/8  │ -                        │ 2d        │
╰───┴─────────────┴─────────┴────────┴───────┴──────────────┴────────────┴──────┴──────────────────────────┴───────────╯
firecracker versions differ across hosts (v1.16.0 on prod-west-1; v1.17.0 on prod-east-1); live migration needs both hosts on the same release, run fc-update on the hosts behind

The * marks the host currently selected as active for this context.

GPUS shows allocated/capacity with the GPU model in parentheses when the host advertises one. A host with no GPU capacity renders -, so a CPU-only fleet stays readable. GPU capacity is only probed and reported on qemu hosts, a firecracker host always shows -. For per-device detail and MIG instance pools, see fuse host get and fuse host metrics.

FIRECRACKER is the firecracker release the host agent reported at registration. When it differs from the release that host’s fc-agent.py pins (FIRECRACKER_VERSION), the pin follows in parentheses: that host has not run fc-update since the pin moved. A qemu host, an agent too old to report a version, or a host registered before the orchestrator stored it shows -. Nothing re-probes an existing host, so after upgrading the orchestrator every host that was already registered shows - until it is registered again. When the listed hosts run more than one release, a warning on stderr names each release and the hosts on it, since live migration needs both hosts on the same release. Only the listed page is compared, pass --all-pages to check the whole fleet. Like LAST SEEN, the version is read once at registration and not refreshed, so it can lag a host that ran fc-update after it registered.

LAST SEEN is stamped at registration and never refreshed, it is not a liveness signal and the scheduler never consults it. Judge reachability from the STATE column and the host agent’s own health endpoint instead.

Select

fuse host [id]

With an ID, validates the host exists (a GET against it) and, if so, sets it as the active host for the current context, persisted to the config file. With no ID, prints the current selection instead of changing it.

fuse host prod-east-1
# selected host "prod-east-1"

fuse host
# active host: prod-east-1

Subcommands (register, get, cordon, uncordon, remove, metrics) take precedence over selection, fuse host register ... runs the register subcommand, not a selection attempt on a host literally named register. They are registered under the plural fuse hosts as well, so neither name is a dead end.